Hi Kevin,
Thank you for your attention!
On the principle, since all parts are cross-platform, it should be installable on all kind of OS: Windows, Linux, Mac (Intel/Silicon), etc. Of course, not all tested on my side, thus certainly some tuning are to be done to succeed with all parts of the code.
My own demo server is Linux (a few technical informations are available on the help hint on top-right of the interface).
The Front/Middleware side is Python. The Back side may be Python or Java (currently Java, much more efficient for the crossword builder algo).
The semantic part (similarity search, thematic glossary creation) is a Qdrant vector database using its ready to use Docker, and a small Qwen3 0.6B embedder (CPU via LlamaCpp).
For the data preprocessing (no need to re-evaluate for a new installation): word statistics were done on a large opus.nlpl.eu corpus set, word/multi-word were enriched with Scrabble dictionaries, Hunspell (also used for its grammatical typologies), and Wiktionary (also used for its definitions). Usage scoring of words and multi-words was done using a local small Qwen3.5 9B LLM (2x3060 12G GPUs using vLLM).
Clue helper is using Mistral Small.
The ChatBot is also using the local LLM.
I would really enjoy all kind of feedback/exchanges!
